When there is a cheater, one has nothing to do but wait. After two minutes my opponent passed again, and I had to respond, this continues until one of us decides to give up. To win such a game requires a considerable amount of unreasonable waiting.
The bots are getting so much better now, why not allowing them to decide a game? If a game had reached the endgame stage, shouldn’t a win rate of >99% be taken as a definite win? This should easily discourage the cheaters.
Edit: Just to clarify. Most of the cases the abusers just refuses to accept the score, aiming to induce a timeout. In those cases, I would be happy to capture the dead stones such that the bot would be able to understand that I am winning. Of course, capturing the dead stones are point-losing moves in the Japanese role. But in most cases, the point differences are so large that I wouldn’t mind doing so.
Of course, this would not be able to solve all the abusive case. But it should certainly discourage the routine abusers (which I encountered many).
I think Anoek mentioned working on something similar, but I am afraid it is not as simple. Simulating perfect play is still rather comupitng power costly and doing it for several games at once might be not as simple.
Also, especially beginners often leave unsettled positions both thinking there is nothing more to gain. They might get very confused when their games are suddenly scored different from they thought… But at least it would be educational
As far as cheaters are concerned I recommend just resigning and reporting
Report the cheater (use the Report Game function) and block them from accepting your open challenges. The mods can always award you back any ranking points that you lose from a resignation anyway.
Glad to know that this is in progress. This would make OGS the most technically advanced online go platform.
Some software just evaluates the winning rate through evaluating the board position, where no simulation is needed. I think the computational cost should be feasible.
typically, they are casual cheaters that keep renewing their accounts. those players just register new accounts and play a few games, until they get banned, then continue with another account. their profiles look surprisingly similar, with very few games played and get lots of complaints from each game.
Makes me think if we can trap cheaters in a honeypot, where cheaters think they are successful in their cheating but really they are just stuck in a sandbox fighting against bots who don’t mind the cheating.
Shadowbanning’s a dark route to go down, I don’t recommend it.
I agree. Not only beginners make that mistake
I won’ game (as an SDK) vs a DDK because we both didn’t see that one of my group is in atari (see game below)
Another problem with that system is… judging from those Leela games, it’s very rare for bot to predict >99% win rate even when it seem like a definite win from human perspective. You have to actually capture dead group to raise the win rate (which is not necessarily possible under Japanese rule). If you settle for something lower, like 95%, it’s a worse idea because highly probably (even for pros) to turn a almost definite win into a loss or the reverse. Your opponent might abuse this this function to force to end a game in progress.
I think abusers are quite rare, so don’t give them more tools. Even I might be tempted to use the force end game function on people I know as a joke. I’ve only met like 2-3 abusers in 2 years as far as I can recall.
TBH, I’m not even sure shadowbanning would even work. Wouldn’t the “not being able to see or accept game offers” be a dead giveaway? Flagging problematic users I think could work. I do believe though, that the focus should be catching these users.
It is probably possible to implement a good autoscore, which would work a lot better than an engine, which is designed to win, or produce a winning percentage, rather than produce an actual score. The trouble with Japanese rules can be easily solved if after a resume, we enter a “scoring dispute” mode where further moves don’t count towards point total (or, alternately, do like AGA rules are start using pass stones, making sure each player eventually plays the same number of moves). Perhaps it would be also possible to determine algorithmically if the game has actually ended or not.
I don’t really expect this to cover all the edge cases though, so in the case of a unresolvable score dispute, there really is no way except adding the game to some sort of mod queue.
There can be a cheater in OGS? How is this possible?
Most of the time the point difference is huge. Abusers just refuse to accept the score aiming to induce a timeout. I wouldn’t mind capturing these dead stones in those cases.
My experience is on 1313 and 99 there are more abusers.
In 9x9 and 13x13, it always seems to be that the opponent if losing tries loads of desperation moves, all of which are obviously refuted and much of which aren’t even sente. Not the topic of the thread but half the small board players do this.
Quite true. This is most likely because inexperienced players play more frequently on 9x9 (and somewhat less on 13x13) rather than “taking the plunge” and playing on 19x19. It’s understandable, but can be frustrating for players with more experience who have these people as opponents. But don’t forget: all of us were beginners at the start of our Go playing days.
Another reason why such a system is necessary is we need to deal with stone fillers.
This player is 2D, but he is doing nothing but filling stone, to the point that I lost my patience and resigned the game. The mods have not reached yet. But if I had made all my groups alive, it should be easy to force a bot decision.
Fantasishada is new and is a troll (see his first few games). His rank was achieved because several players, including a 1d, resigned just as you did when confronted with his antics. This drove his rank from the default start of 13k to 2d, though he has now fallen back to 8k. He should be reported and banned. Thank you for bringing him to everyone’s attention.
This topic was automatically closed 91 days after the last reply. New replies are no longer allowed.